Haze Removal via Constrained Transmission Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing haze removal techniques often produce images with artifacts such as noise boosting, excessive de-hazing, and incorrect color shifts, making them unsuitable for various image processing applications.
Innovation Solution
The technique constrains medium transmission in the haze image formation model to be greater than or equal to a fractionally scaled variant of the input image, filters it for local smoothness with sharp discontinuities along image edges, and converts images to gamma decoded sRGB linear space for haze removal, followed by gamma encoding, to preserve scene depth and reduce ambiguity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing haze removal techniques are applied, then haze is removed from the image, but artifacts such as noise boosting, excessive de-hazing, and incorrect color shifts are produced
Solution Approach 1:
The patent applies preliminary constraints to the transmission map before haze removal processing. Specifically, it enforces geometric constraints (0 ≤ t(x) ≤ 1), brightness constraints (0 ≤ J(x) ≤ 1), and content visibility constraints (J(x) ≥ s·I(x)) to prevent artifacts from forming in the first place. This preliminary anti-action approach stops noise boosting, excessive de-hazing, and color shifts before they occur by bounding the transmission values and recovered scene radiance within valid ranges.
Solution Approach 2:
The patent employs an iterative optimization process where the transmission map is refined through multiple passes. In each iteration, the algorithm computes the recovered scene radiance, compares it against the constraints, and adjusts the transmission map accordingly. This feedback loop continues until convergence, ensuring that the final transmission map satisfies all constraints while producing artifact-free haze removal results.
2Illumination intensity
If strong haze removal is applied to improve visibility, then haze is reduced, but image regions become overly dark
Solution Approach 1:
The patent prevents overly dark regions by applying the content visibility constraint J(x) ≥ s·I(x) where s is a scaling factor (e.g., 0.1). This constraint ensures that the recovered scene radiance never falls below a fraction of the input hazy image radiance, thereby maintaining minimum brightness levels and preventing complete darkening of image regions while still removing haze effectively.
3Stability of the object's composition
If transmission filtering is applied to preserve scene depth, then local smoothness is achieved, but computational complexity increases
Solution Approach 1:
The patent performs transmission filtering as a preliminary step before the main haze removal optimization. By pre-filtering the transmission map to enforce local smoothness with sharp discontinuities at edges (using methods like guided filtering or bilateral filtering), the algorithm reduces the search space for the optimization process and prevents high-frequency noise in the transmission map from causing artifacts. This preliminary action simplifies the subsequent optimization while preserving scene depth.
4Reliability
If atmospheric light estimation is performed to remove haze, then haze removal is achieved, but blue cast artifacts occur due to incorrect color shifts
Solution Approach 1:
The patent prevents blue cast artifacts by applying geometric constraints (0 ≤ t(x) ≤ 1) and brightness constraints (0 ≤ J(x) ≤ 1) to the transmission map and recovered scene radiance. These constraints ensure that the atmospheric light estimation and subsequent haze removal operations remain within physically valid ranges, preventing color channel imbalances that would manifest as blue casts. The constraints act as a safeguard against incorrect color shifts throughout the processing pipeline.
Data Source
AI summary
Techniques are disclosed for removing haze from an image or video by constraining the medium transmission used in a haze image formation model. In particular, a de-hazed scene, which is a function of a medium transmission, is constrained to be greater than or equal to a fractionally scaled variant of the input image. The degree to which the input image is scaled can be selected manually or by using machine learning techniques on a pixel-by-pixel basis to achieve visually pleasing results. Next, the constrained medium transmission is filtered to be locally smooth with sharp discontinuities along image edge boundaries to preserve scene depth. This filtering results in a prior probability distribution that can be used for haze removal in an image or video frame. The input image is converted to gamma decoded sRGB linear space prior to haze removal, and gamma encoded into sRGB space after haze removal.